Microsoft Word - scribe_1_.doc
|
|
- 巫躬 明
- 7 years ago
- Views:
Transcription
1 Making Faces 2005/06/08 R 陳 坤 毅 R 莊 曜 誠 R 王 博 民 3D acquisition for faces: 如 何 取 得 臉 的 3D model? 方 法 1: ( 經 費 足 夠 時 ) 使 用 Cyberware scanner. ( 可 對 臉 部 scan 亦 可 對 全 身 scan) 方 法 如 下 圖 所 示 : 缺 點 : 人 必 須 保 持 靜 止 無 法 取 得 動 態 的 3D model, 且 成 本 較 高 優 點 : 比 較 能 得 到 精 確 的 Mode 方 法 2: ( 由 照 片 取 得 3D model.) 方 法 概 念 與 mosaic 類 似, 先 給 一 個 人 臉 形 狀 的 model ( 積 木 ), 用 warping morphing 或 是 調 整 一 些 參 數, 使 得 由 各 個 角 度 去 看 時, feature points 能 夠 match 變 型 後 的 人 臉 好 處 : 人 可 以 動, 不 須 需 靜 止
2 Step1: 用 五 台 camera 拍 攝 人 臉 Step2: User 必 須 給 定 幾 十 個 feature point, 指 出 各 個 部 位 的 位 置 Step3: 估 計 feature point 3D 中 座 標 以 及 Camera 的 參 數 ( 亦 即 Structure from motion) Step4: 用 更 複 雜 的 feature points 把 model 再 更 進 一 步 deform, 使 能 更 match 臉 上 的 一 些 feature Step5: 最 後 將 texture 貼 上 Mesh deformation 1. 計 算 feature point 該 移 動 到 哪 裡 去 2. 決 定 那 些 並 非 對 應 到 feature point 的 vertex 必 須 對 應 到 哪 邊 去 Scattered data interpolation 黃 色 點 為 feature point, 灰 色 點 為 整 個 model 重 心 中 間 圖 白 色 位 置 為 feature point 原 來 位 置, 黃 色 為 白 點 將 被 移 動 到 的 地 方 欲 決 定 黑 邊 上 的 點 要 移 動 到 哪 去, 先 去 看 週 圍 的 點 移 動 到 哪 去, 再 用 一 個 smooth 的 kernel 來 做 一 個 weighted average. Texture extraction
3 分 為 view independent and view-dependent, 但 view dependent 通 常 提 供 較 好 的 texture 做 weighting 前 必 須 要 考 慮 到 以 下 幾 點 : 1. Occlusion: ( 某 些 部 位 只 有 在 特 定 幾 張 image 出 現 ) 2. Smoothess: 轉 動 時 不 要 跳 動 3. Positional certainty: normal 越 朝 像 中 間 越 準 確, 越 朝 邊 邊 越 不 準 確 4. View similarity: 越 靠 近 此 view 的 weighting 會 越 高 View-independent VS. View-dependent View-independent: 快, 但 會 有 blur 的 現 象 View-dependent: 慢, 但 feature 會 比 較 像 Model reconstruction: Input: 給 不 同 時 間 點 五 張 照 片, 一 個 以 及 generic 3D model Output: 該 時 間 點 的 3D model 以 及 view-dependent texture Creating new expression: Global blending: ( 每 個 點 都 apply 相 同 的 weighting) 下 面 是 一 個 簡 單 的 例 子
4 左 圖 跟 中 間 的 圖 都 各 用 一 半 的 weighting, 最 後 再 做 blending, 可 得 到 右 圖 這 張 新 的 Expression Region-based blend: 下 面 是 一 個 簡 單 的 例 子 我 們 取 左 邊 的 上 半 部 份 臉 部, 和 中 間 的 下 半 部 份 臉 部, 可 得 到 右 圖 新 的 expression Painterly interface:
5 取 各 張 image 我 們 所 要 的 部 份, 合 成 出 一 張 新 的 expression Animating between expressions 以 下 面 的 例 子 說 明 : Input: 一 張 自 然 的 表 情, 以 及 一 張 開 心 的 表 情 我 們 可 以 利 用 morphing 的 方 式 將 中 間 的 表 情 建 構 出 來 Spacetime faces: 建 立 3Dmodel 的 另 一 種 方 法 ( 發 表 於 2003 年 ), 能 表 現 出 更 detail 的 部 份 ( 如 皺 紋 ) ( 有 別 於 前 述 發 表 於 1997 年 的 方 法 ) 會 用 到 下 圖 設 備, 使 用 Stereo 的 方 式 建 立 3D model
6 其 中 : Black & white camera 用 來 補 捉 range image Color Camera 用 來 補 捉 texture Video projectors 會 投 影 出 一 些 patter 到 人 的 臉 上 這 些 Cameras 均 為 60 frames/sec 3CCD 640*480 的 Resolution 下 圖 為 各 個 camera 在 不 同 時 間 點 所 得 到 的 不 同 影 像 每 三 個 frames 為 一 個 單 位, 兩 個 有 投 影 pattern 到 臉 上, 一 個 沒 有 投 影 pattern 到 臉 上
7 而 我 們 的 目 的 是 決 定 出 與 沒 有 投 影 的 那 個 frame match 的 3D model Stereo: 單 純 地 直 接 對 該 frame 做 Stereo analysis:
8 找 出 該 frame 中 每 個 點 會 形 成 一 個 block, 利 用 block match 的 方 式 找 到 另 一 台 像 機 該 frame 的 對 應 點 如 此 一 來 便 可 由 2D image 利 用 stereo 的 方 式 找 出 3D mode 如 下 圖 所 示 由 於 在 做 block matching 時 會 有 一 些 ambiguous 的 情 況 發 生, 所 以 結 果 不 會 太
9 好, 我 們 發 現 臉 部 會 有 許 多 noise 的 現 象, 因 此 我 們 需 要 用 投 影 pattern 的 方 式 來 改 善 如 下 圖 所 示 Active Stereo: 我 們 會 利 用 投 影 的 pattern 來 找 block 的 match, 以 減 輕 誤 差 我 們 發 現 結 果 會 好 很 多 但 結 果 仍 不 太 滿 意, 因 此 我 們 用 spacetime stereo, 方 法 如 下 : Spacetime stereo:
10 一 次 考 慮 5 個 frame, 使 效 果 更 好 Temporally 來 看, 五 個 frame 中 的 block 都 要 match Ex. 8*8 的 block, 則 是 8*8*5 的 volume 在 5 個 frame 都 要 match 但 我 們 必 須 要 克 服 掉 motion 的 問 題 我 們 會 假 定 這 一 個 volume 只 是 一 個 "orthogonal volume 做 一 個 affine transformation" ( 亦 即 臉 的 motion 為 linear 移 動 ) 第 一 個 frame 的 對 應 :
11 第 二 個 frame 的 對 應 : 第 三 個 frame 的 對 應 :
12 第 四 個 frame 的 對 應 :
13 第 五 個 frame 的 對 應 : 找 出 對 於 整 個 volume 最 match 的 點
14 示 意 圖 : Spacetime Stereo 的 缺 點 : 在 這 樣 的 device 下, 人 通 常 不 太 願 意 把 眼 睛 張 開, 因 為 兩 個 projector 直 射 在 臉 上 Fitting: 將 左 邊 與 右 邊 的 range image 合 起 來 3D face applications : 以 下 為 三 部 電 影 用 到 3D face 的 例 子 : 1. The One ( 救 世 主 )
15 劇 情 為 兩 個 相 同 的 人 對 打, 左 圖 為 替 身, 袋 上 綠 色 面 具, 我 們 將 3D model 覆 上 即 可 2. Gladiator ( 拍 攝 此 電 影 時, 該 角 色 人 物 死 亡, 後 來 的 部 份 均 以 3D face 解 決 )
16 3. Spiderman 2 這 整 個 場 景 均 為 CG 所 產 生, 缺 點 在 於 此 人 的 臉 部 完 全 無 表 情 變 化 (CYY 按 : 因 為 他 快 死 了 ) Statistical methods 是 用 來 解 決 ill-posed 的 問 題 Ill-posed:
17 y 給 定 一 些 parameter z, 帶 入 function, 再 加 上 noise, 得 到 observed signal 要 解 決 的 問 題 是 : 如 何 從 所 得 到 的 y,recover 回 到 z (f 為 未 知 ) 例 子 : (1)super-resolution: input image (parameter) 為 一 個 high resolution image, 例 如 為 128*128 的 image, 經 過 down-sampling( 使 用 一 個 4*4 的 box, 將 box 內 的 值 取 平 均 作 sample 的 值 ), 則 我 們 所 觀 察 到 的 image 為 64*64 的 image 所 以 這 裡 的 ill-pose 是 如 何 從 64*64 的 observed signal 來 recover 回 128*128 的 parameter 雖 然 我 們 知 道 forward process, 但 是 我 們 所 要 的 parameter 個 數 比 我 們 所 觀 察 到 的 個 數 還 要 多 (2)de-nosing: input 為 原 來 的 image,f 為 jpeg compression operation (DCT quantization), 得 到 一 張 compressed image, 我 們 所 觀 察 到 的 就 是 一 張 compressed image, 雖 然 這 邊 二 邊 的 image 的 dimension 一 樣, 但 是 經 過 compressed 後 很 多 detail 也 都 不 見 了, 所 以 這 邊 也 算 是 ill-pose 的 問 題 ; 我 們 如 何 從 一 張 high compressed image 來 recover 回 原 來 的 image 所 以 statistical method 是 用 來 解 決, 如 何 從 observed signal 來 recover 回 原 來 的 parameter 找 一 個 optimal parameter z, 使 得 P( z y) 最 高 以 super-resolution 為 例, 給 定 一 張 64*64 的 image y, 要 找 出 一 張 128 個 128 的 image z, 使 得 P( y z) P( z) P( z y) 為 最 高 根 據 貝 式 定 理 知 道 P ( z y) =, 因 為 對 z 做 P( y)
18 quantization, 所 以 P(y) 跟 optimalize 無 關, 可 以 把 它 當 作 constant 忽 略 掉, 再 取 log 變 成 likelihood, 則 乘 法 變 加 法, 因 此 得 到 minimize * L ( y z) + L( z) 所 以 整 個 方 程 式 可 以 寫 成 z = min L( y z) + L( z) z Data evidence and prior knowledge: 我 們 要 猜 一 組 z, 代 入 f, 使 得 f(z) 和 y 最 接 近 以 super solution 為 例, 如 果 觀 察 到 的 y 是 64*64 的 灰 色 的 圖 (intensity level 都 是 128), 則 我 們 會 猜 z 可 能 是 128*128 的 灰 色 image, 也 可 能 會 猜 z 是 128*128 的 chess board pattern( 黑 白 相 間 ) 的 image 所 以 我 們 必 須 使 用 prior knowledge 讓 我 們 的 ill-posed problem 變 成 well condition problem 所 以 如 果 我 們 知 道 原 來 的 image 是 smooth image, 則 我 們 就 可 以 知 道 z 是 128*128 的 灰 色 image 而 不 是 chess board pattern 的 image 8 如 果 給 一 個 10*10 的 灰 階 的 block image, 每 個 pixel 有 2 種 顏 色, 則 這 樣 的 10*10 的 block 有 10 種 變 化, 如 果 每 一 個 人 一 天 大 約 看 8* 10 blocks, 22 全 球 有 60 億 人, 則 全 球 每 天 看 的 image block 約 10, 所 以 要 多 少 人 多 少 年 才 能 把 所 有 的 image block 看 完 意 思 是 說,image 出 現 的 頻 率 並 不 是 equal 的, 有 些 image 可 能 ㄧ 輩 子 都 看 不 到, 有 些 image 出 現 頻 率 會 比 其 他 高 很 多, 所 以 我 們 傾 向 於 相 信 有 些 image 是 好 的 image( 就 是 我 們 想 要 的 image), 有 些 是 壞 的 image( 大 致 都 不 會 出 現 的 image), 所 以 我 們 才 用 statistical method
19 Generic priors: 所 以 我 們 知 道 現 在 有 好 的 image 和 壞 的 image, 則 我 們 如 何 來 encode 好 的 image, 最 常 用 的 就 是 generic priors 因 為 什 麼 都 不 知 道, 所 以 先 假 設 smooth 的 image 就 是 好 的 image, 想 辦 法 去 encode z,v(x) 去 考 慮 和 鄰 居 間 的 frequency, 通 常 我 們 會 用 Gaussian MRF 和 Huber MRF( 比 較 robust) 但 是 generic prior 得 到 的 結 果 不 是 很 好, 比 較 模 糊 Example-based priors:
20 假 設 有 出 現 過 的 image 就 是 好 的 image, 則 怎 樣 encode 一 張 曾 經 出 現 過 的 image 就 是 好 的 image, 在 super-resolution 的 應 用 : 去 收 集 6 張 200*200images, 當 中 只 要 出 現 過 的 image block 就 是 好 的 image block, 所 以 去 生 成 image 的 時 候, 那 些 block 最 好 是 出 現 過 的 以 super-resolution 為 例 子, 先 給 low resolution 的 blocks, 利 用 六 張 的 200*200 的 images 我 都 可 以 down-sampling 得 到 low-resolution, 所 以 我 們 得 到 high-resolution 對 low-resolution 的 care, 我 們 就 去 找 跟 我 們 low resolution 的 block 比 較 像 的, 然 後 將 所 對 應 的 high-resolution 當 做 這 塊 的 high-resolution, 如 果 是 一 塊 一 塊 去 生 成, 必 須 要 考 慮 overlap 的 問 題, 其 overlap 也 要 越 相 近 越 好
21 Model-based priors: 如 果 現 在 要 處 理 人 臉 的 image, 那 人 臉 的 image 就 是 好 的 image, 狗 的 貓 的 樹 的 image 就 不 是 好 的 images Model-based prior 基 本 上 說 你 有 一 個 club, 將 所 有 屬 於 這 個 club 的 image 產 生 一 個 model, 來 summarize 這 些 image, 最 常 用 的 就 是 Gaussian model Gaussian model 基 本 上 是 對 這 些 images 做 PCA 如 果 我 們 知 道 這 張 是 人 臉 的 照 片,L(z) 是 encode 怎 樣 的 image 才 算 是 人 臉 的 image, 基 本 上 會 從 一 堆 人 臉 的 training data 求 出 parametric model, 然 後 利 用 parametric model 來 evaluate 任 何 的 input image 是 否 為 人 臉 的 image 我 們 用 PCA 來 做 這 件 事 情, 我 們 可 以 想 像 所 有 的 image 都 是 high dimensional vector, 然 後 PCA 會 把 它 project 到 low dimensional 的 space 上 去, 可 是 這 low dimensional space 還 是 可 以 很 真 實 的 表 現 出 我 原 來 的 data, 所 以 可 以 像 是 我 用 比 較 compact 的 技 術 來 表 現 比 較 複 雜 的 image, 所 以 每 個 人 臉 的 image 都 可 以 用 比 較 少 量 的 技 術 S 去 生 成,Z 是 人 臉 的 image, S 就 是 這 些 技 術 (low dimensional), 例 如 我 們 將 1000 張 的 128*128 的 image (Z) 作 PCA, 可 以 reduce 到 20 dimensional 的 space, 則 S 就 是 20 dimensional 的 vector, 每 個 20 dimensional 的 image 就 可 以 生 成 類 似 人 臉 的 image, 所
22 以 只 要 有 S 就 可 以 生 出 Z, 在 作 PCA 的 過 程, 同 時 也 把 這 些 20 個 dimension 的 每 一 個 component 可 能 出 現 的 機 率 都 有 prior model, 所 以 現 在 換 成 我 的 prior model 是 對 技 術 S 的 prior model, 一 但 有 S 就 可 以 生 出 Z, 所 以 我 們 基 本 上 把 Z 換 成 S, 所 以 整 個 是 對 S 來 做 Z 可 以 寫 成 S 的 一 些 combination PCA: Principal Components Analysis (PCA) 以 2d 為 例 子, 給 ㄧ 堆 綠 色 的 data points, 如 果 我 們 用 原 來 的 座 標 軸 來 描 述 他, 會 發 現 不 是 exactly 的 描 述 方 式, 我 們 想 用 更 compact 的 方 式 來 描 述 他 這 2d 的 data, 我 們 希 望 用 1d 的 技 術 來 描 述, 把 這 些 點 投 射 到 1d 上, 取 任 何 一 軸 都 無 法 很 真 實 的 呈 現 這 些 data 那 我 們 應 該 怎 麼 找 座 標 軸, 經 過 projection 後 還 能 很 真 實 的 呈 現 原 來 的 data 最 簡 單 的 方 式 是 先 找 一 個 軸, 使 得 其 他 點 投 影 到 這 軸 的 variance 最 大, 接 下 來 找 variance 第 二 大 的 且 跟 原 來 的 軸 垂 直, 然 後 我 們 就 將 綠 色 的 點 轉 到 first principal component 和 second principal component 所 描 述 的 座 標 系 裡 面, 我 們 可 以 發 現 這 座 標 系 能 更 真 實 的 呈 現 這 些 datas 所 以 如 果 在 n-dimension 系 統, 做 過 PCA 之 後, 我 們 只 要 把 variance 小 的 軸 都 拿 掉, 就 可 以 將 data 做 到 dimension reduction 的 動 作 可 以 把 它 想 成 是 一 個 Gaussian function, 把 這 些 點 算 出 一 個 Gaussian function 來 描 述 這 些 2d 的 點, 所 以 作 PCA 跟 做 Gaussian function 基 本 上 是 一 樣 的 下 面 是 作 PCA 的 步 驟 :
23 現 在 要 做 人 臉 的 image, 那 怎 樣 的 image 是 好 的 image, 先 給 training data, 我 們 求 出 這 些 image 的 mean vector, 就 是 上 圖 的 Average face, 我 們 用 PCA 算 出 principal component, 我 們 取 15 個 principal component, 所 以 對 所 有 128*128 的 image 我 們 都 可 以 用 這 15 個 技 術 來 表 示, 所 以 一 張 圖 如 果 是 A1~A15 的 技 術 所 組 成, 則 此 張 圖 的 表 示 方 式 是 mean + A1*( 第 1 個 component) + A2*( 第 2 個 component) + + A15*( 第 15 個 component), 這 這 張 image 就 可 以 很 接 近 原 來 輸 入 的 image 為 什 麼 取 15, 是 因 為 我 們 觀 察 energy decade, 發 現 到 某 個 值 之 後 variance 都 很 小, 意 思 是 說 這 幾 個 feature component 沒 有 什 麼 太 大 的 影 響, 也 就 是 說 本 來 128*128 的 image, 本 身 dimension 就 沒 那 麼 高, 因 為 之 間 有 強 烈 的 coherence, 所 以 用 15 張 就 可 以 cover 所 有 的 variance, 所 以 PCA 的 意 思 就 是 這 樣 上 圖 是 對 人 臉 做 super resolution, 然 後 使 用 三 種 priors 的 結 果 比 較,(a) 為 input low resolution image,(f) 為 真 正 的 high resolution image, 基 本 上 是 每 個 方 法 將 low resolution 求 出 自 己 認 為 的 high resolution image (c) 是 用 generic prior 做 的 結 果, 所 以 得 到 的 結 果 比 較 模 糊,(d)(e) 是 example-based prior 得 到 的 結 果, 但 是 其 結 果, 有 的 時 候 人 臉 的 地 方 不
24 像 人 臉, 例 如 鼻 子 的 地 方 有 很 奇 怪 的 鼻 線 之 類 的, 因 為 它 只 考 慮 local 的 face 沒 有 考 慮 global 的 部 份,(b) 是 model-based prior, 基 本 上 它 是 model-based 加 上 example-based 在 global 部 分, 他 用 PCA 算 出 人 臉 大 概 應 該 長 什 麼 樣 子, 所 以 產 生 出 來 的 image ㄧ 定 要 長 的 像 人 臉, 不 能 脫 離 人 臉 太 遠, 可 是 在 local 的 face 他 用 example-face, 所 以 結 果 比 其 他 三 張 的 結 果 更 像 人 臉 Face models from single images 用 ranging scanner 去 scan 200 個 ranging image,100 個 男 生 100 個 女 生, 同 時 取 得 其 texture map, 以 此 200 個 ranging images 建 立 一 個 prior model, 每 一 張 ranging image 都 是 一 堆 features 的 集 合, 例 如 眼 角, 嘴 角 這 些 明 顯 的 地 方, 此 model 即 為 morphable 的 face model 接 著 input 一 個 2D 的 image, 我 們 就 利 用 這 個 morphable face model, 用 上 面 那 兩 百 個 model 以 2D features 可 以 找 得 到 對 應 點, 以 及 可 以 產 生 出 一 個 完 整 的 mesh 的 情 況 下, 即 可 產 生 一 個 face model, 其 可 以 match 我 們 的 input image, 且 是 一 個 完 整 的 mesh 然 後 我 們 可 以 對 此 face mesh 做 一 些 morphing 的 變 化, 使 用 此 model 的 一 些 gain, ranging image, image texture, 用 PCA 去 求 出 一 個 該 face shape 的 prior model, 我 們 即 可 用 此 方 法 來 調 整 一 些 參 數, 使 得 我 們 的 face model 可 以 做 出 一 些 表 情, 甚 至 可 以 讓 這 個 人 變 胖 變 瘦 等 像 是 整 型 的 改 變
25 接 下 來 講 到 有 關 數 學 理 論 的 部 分 使 用 我 們 的 200 張 ranging images, 將 每 張 image 都 align 好, 每 一 個 相 對 應 的 feature 都 互 相 對 應 好, 每 一 個 ranging image 就 變 成 了 vertex 所 形 成 的 集 合, 有 n 個 vertex 就 是 n dimensional or 3n dimensional( 彩 色 的 話 ) 的 vector, 所 以 總 共 就 有 200 個 images, 每 個 image 的 dimension 是 3n, 所 以 我 們 可 以 將 他 看 成 是 有 200 個 3n dimensional 的 points, 我 們 即 可 用 PCA 的 方 式 去 算 出 這 200 個 points 的 mean 以 及 principal component, 假 設 我 們 取 其 中 的 m 個 components, 我 們 有 了 mean face shape 和 m 個 face shape 的 principal component, 所 以 每 一 個 face mesh 都 可 以 投 影 到 這 個 m dimension 的 vector 上 面 去, 所 以 只 要 給 你 一 個 m dimensional 的 vector, 我 們 都 可 以 求 出 其 對 應 的 ranging image 每 一 個 face image 都 是 一 個 md 的 vector, 係 數 即 為 α1 到 αm,si 就 是 第 i 個 shape 的 principal component,αi 就 是 對 應 到 該 component 的 coefficient, 而 βi 代 表 其 相 對 應 principal component 的 texture 的 coefficient, 而 且 α 有 他 自 己 的 probability function, 之 前 我 們 有 用 SVD 算 出 其 singular value, 而 singular value 所 對 應 的 就 是 他 的 covariance, 所 以 對 每 個 component 我 們 知 道 他 的 係 數 α
26 i 的 分 佈 可 已 有 多 廣, 對 於 每 一 個 α, 我 們 都 可 以 得 到 其 prior model, 這 就 是 我 們 要 的 morphable model, 以 此 便 能 對 於 我 們 所 input 的 face model 加 以 變 化 如 果 我 們 先 以 四 個 features 為 例 的 話,Divide face into 4 regions (eyes, nose, mouth, head),for each new prototype, find amount of deviation from the reference shape and texture 我 們 可 以 在 這 四 個 features 上 標 上 一 些 特 性, 再 以 PCA 的 方 法 來 產 生 face model 下 面 這 個 奧 黛 麗 赫 本 是 另 外 一 個 例 子,input 一 張 她 的 2D image, 就 可 以 重 建 她 的 3D model, 然 後 我 們 就 可 以 任 意 的 對 這 個 3D model 做 一 些 變 化, 像 是 加 上 帽 子, 或 是 笑 得 更 開 一 些
27 下 面 這 個 例 子 是 將 兩 個 真 實 世 界 裡 的 人 臉, 套 用 在 電 影 海 報 上 面 或 是 畫 像 上 面, 讓 人 們 真 的 可 已 變 成 畫 裡 面 的 男 女 主 角 再 來 這 個 應 用 是 用 在 髮 型 上 的 變 化, 如 果 你 不 知 道 自 己 在 什 麼 樣 的 髮 型 下 面 會 變
28 成 什 麼 樣 子, 或 是 想 要 預 覽 一 下 自 己 剪 完 頭 髮 之 後 的 樣 子, 就 可 以 使 用 下 面 這 個 技 巧 最 後 一 個 應 用 是 在 人 的 體 型 上 面 的 應 用, 我 們 可 以 用 相 同 的 方 法, 對 於 人 的 體 型 也 建 立 出 一 個 morphable 的 body model, 來 模 擬 我 們 所 要 呈 現 的 body model, 如 此 一 來, 我 們 便 能 用 一 些 簡 單 的 參 數 來 控 制 高 矮 胖 瘦 以 及 一 些 身 體 的 特 徵, 或 是 做 出 一 些 特 別 的 動 作
29 Image-based faces (lip sync.) 有 了 之 前 我 們 討 論 的 3D mesh 之 後, 我 們 想 要 讓 一 個 人 講 出 他 沒 有 說 過 的 話 首 先 input 一 個 training video, 將 此 video 分 成 影 像 跟 聲 音 兩 部 分, 對 語 音 做 segmentation 將 語 音 分 成 很 多 個 音 節, 並 且 將 其 音 節 跟 其 對 應 的 video 記 錄 下 來, 接 著 input 一 段 新 的 語 音, 也 將 此 新 的 語 音 分 成 多 個 音 節, 對 於 每 個 新 的 音 節, 我 們 到 之 前 建 立 的 database 裡 去 找 出 最 像 的 音 節, 然 後 將 此 音 節 所 對 應 到 的 那 一 小 段 video sequence 找 出 來, 將 這 些 小 的 video 連 接 起 來, 就 可 以 得 到 我 們 的 output sequence
30 Analysis stage Video Model Background Video Synthesis stage Select Lip Video Stitch Together 因 為 人 是 會 動 的, 所 以 還 要 加 上 一 些 global motion estimation, 像 是 頭 整 個 會 動 嘴 巴 也 要 跟 著 動, 整 個 看 起 來 會 更 自 然 一 些 而 且 人 們 對 於 語 音 跟 嘴 巴 的 對 應 其 實 容 忍 度 還 蠻 高 的, 像 是 有 些 韓 劇 用 中 文 配 音 我 們 也 覺 得 還 蠻 搭 的, 所 以 這 個 research 有 些 時 候 input video sequence 不 是 很 長,sample 沒 有 很 多, 不 過 整 體 看 起 來 會 覺 得 也 都 還 蠻 順 的 Relighting faces face relighting 的 原 理 很 簡 單, 因 為 光 是 可 以 相 加 的, 像 是 下 圖, 如 果 你 將 左 邊 的 燈 亮 起 來 會 得 到 第 一 張 照 片, 如 果 將 有 邊 的 燈 亮 起 來 會 得 到 第 二 張 照 片, 於 是 如 果 要 同 時 將 兩 盞 燈 都 亮 起 來, 其 實 很 簡 單, 就 是 直 接 將 這 兩 張 照 片 相 加 即 可, 如 果 光 源 是 有 顏 色 的 光, 其 加 成 的 方 法 也 是 一 樣 的
31 接 下 來 這 個 方 法 就 是 建 立 在 上 述 的 性 質 上 面, 我 們 可 以 將 人 臉 或 是 人, 放 到 一 個 我 們 所 想 要 的 場 景 裡 面 操 作 的 過 程 中, 我 們 會 用 到 一 個 device 叫 做 light stage, 在 light stage 上 面 有 三 個 cameras, 人 要 坐 在 裡 面 一 段 時 間 靜 止 不 動, 然 後 camera 會 依 照 不 同 的 光 源 做 各 個 方 向 的 旋 轉, 拍 下 兩 千 多 張 照 片 當 作 database 利 用 這 兩 千 多 張 的 照 片, 我 們 就 可 以 對 任 何 的 場 景 做 relighting, 下 圖 為 這 兩 千 多 張 照 片 的 其 中 幾 張, 照 片 下 方 的 白 點 代 表 的 是 光 源 的 方 向 所 以 我 們 記 錄 了 當 燈 在 某 個 方 向 的 時 候, 整 個 model 的 每 個 位 置 的 顏 色 是 什 麼,
32 下 面 這 張 圖 我 們 可 以 看 到 會 有 一 些 黑 影 (occlusion), 這 是 light stage 本 身 的 鐵 欄 杆 的 影 子, 以 及 白 影 (flare), 這 是 當 光 源 正 射 到 camera 是 所 造 成 的 現 象, 所 以 我 們 要 將 上 述 兩 個 現 象 給 除 去 於 是 我 們 就 可 以 模 擬 我 們 的 model 在 任 何 一 種 光 源 的 場 景 下 面 所 呈 現 的 樣 子, 即 使 是 多 個 不 同 且 複 雜 的 光 源 下 也 沒 問 題
33 最 後 我 們 要 談 的 是 有 關 如 何 測 量 出 現 在 我 們 的 場 景 中 有 哪 些 光 源, 該 如 何 模 擬 這 些 資 料, 我 們 所 會 用 到 的 是 一 個 鏡 面 球 形 的 東 西, 它 可 以 幫 我 們 收 集 環 境 光 源, 好 讓 我 們 方 便 模 擬 這 整 個 場 景
Microsoft Word - KSAE06-S0262.doc
Stereo Vision based Forward Collision Warning and Avoidance System Yunhee LeeByungjoo KimHogi JungPaljoo Yoon Central R&D Center, MANDO Corporation, 413-5, Gomae-Ri, Gibeung-Eub, Youngin-Si, Kyonggi-Do,
More informationMove Component Object selection Component selection UV Maya Hotkeys editor Maya USING MAYA POLYGONAL MODELING 55
3 55 62 63 Move Component 63 70 72 73 73 Object selection Component selection UV Maya Hotkeys editor Maya 55 USING MAYA POLYGONAL MODELING Maya: Essentials Maya Essentials F8 Ctrl F9 Vertex/Face F9 F10
More information(baking powder) 1 ( ) ( ) 1 10g g (two level design, D-optimal) 32 1/2 fraction Two Level Fractional Factorial Design D-Optimal D
( ) 4 1 1 1 145 1 110 1 (baking powder) 1 ( ) ( ) 1 10g 1 1 2.5g 1 1 1 1 60 10 (two level design, D-optimal) 32 1/2 fraction Two Level Fractional Factorial Design D-Optimal Design 1. 60 120 2. 3. 40 10
More informationOutline Speech Signals Processing Dual-Tone Multifrequency Signal Detection 云南大学滇池学院课程 : 数字信号处理 Applications of Digital Signal Processing 2
CHAPTER 10 Applications of Digital Signal Processing Wang Weilian wlwang@ynu.edu.cn School of Information Science and Technology Yunnan University Outline Speech Signals Processing Dual-Tone Multifrequency
More information(Pattern Recognition) 1 1. CCD
********************************* ********************************* (Pattern Recognition) 1 1. CCD 2. 3. 4. 1 ABSTRACT KeywordsMachine Vision, Real Time Inspection, Image Processing The purpose of this
More informationMicrosoft PowerPoint - Sens-Tech WCNDT [兼容模式]
X-ray data acquisition systems for NDT applications 技股份有限公司 先锋科技股份有限公司 科技股份有限公司 先锋科技股份有限公司 www Sens-Tech Ltd UK based company 40 Staff Specialise in detection and data acquisition systems for light and
More informationuntitled
LBS Research and Application of Location Information Management Technology in LBS TP319 10290 UDC LBS Research and Application of Location Information Management Technology in LBS , LBS PDA LBS
More informationθ 1 = φ n -n 2 2 n AR n φ i = 0 1 = a t - θ θ m a t-m 3 3 m MA m 1. 2 ρ k = R k /R 0 5 Akaike ρ k 1 AIC = n ln δ 2
35 2 2012 2 GEOMATICS & SPATIAL INFORMATION TECHNOLOGY Vol. 35 No. 2 Feb. 2012 1 2 3 4 1. 450008 2. 450005 3. 450008 4. 572000 20 J 101 20 ARMA TU196 B 1672-5867 2012 02-0213 - 04 Application of Time Series
More information入 指 令 如 : 鍵 盤 鼠 標 多 點 觸 控 重 力 感 應 陀 螺 儀 等 4. 圖 形 用 戶 接 口 掌 握 引 擎 API 中 GUI 與 GUILayout 類 中 方 法 和 變 量 的 使 用, 能 夠 通 過 GUI Skin 為 應 用 的 界 面 定 制 不 同 風 格 主
Unity 專 業 能 力 認 證 (Unity Certified Professional) 應 考 須 知 壹 閱 讀 大 綱 主 題 一 : 數 字 媒 體 基 礎 知 識 1. 2D 圖 形 知 識 熟 悉 常 用 位 圖 數 據 的 結 構 和 格 式 其 中 包 括 像 素 數 據 顏 色 的 值 與 通 道 的 概 念 2. 3D 圖 形 學 相 關 知 識 a) 理 解 向 量 的
More informationCMOS线性响应测试
应 用 CMOS 高 亮 度 天 文 测 光 CMOS-based High Intensity Photometry in Astronomy 项 目 完 成 人 : 童 仲 毅 俞 雯 萱 田 哲 毅 指 导 老 师 : 林 岚 杭 高 天 文 台 二 〇 一 一 年 八 月 II 摘 要 本 课 题 基 于 人 造 星 点 测 光 实 验 和 验 证 性 天 文 拍 摄 实 验, 旨 在 探
More informationuntitled
TFT-LCD Mura & Y.H. Tseng 2006.12.4 Outline Mura Mura Mura 類 度 Mura Mura JND Mura Convolution filter (Filter design) Statistical method (ANOVA,EWMA) Backgroup estimation (LSD) 2 What is Mura- Mura Mura
More information10384 27720071152270 UDC SHIBOR - Research on Dynamics of Short-term Shibor via Parametric and Nonparametric Models 2 0 1 0 0 5 2 0 1 0 0 5 2 0 1 0 0 5 2010 , 1. 2. Shibor 2006 10 8 2007 1 4 Shibor
More informationSubspace Gradient Domain Mesh Deformation
Subspace Gradient Domain Mesh Deformation Presented by Xinguo Liu( 刘新国 ) State Key Lab of CAD&CG, Zhejiang Univ. To Appear in ACM SIGGRAPH 2006 Jin Huang, Xiaohan Shi, Xinguo Liu, Kun Zhou,Li-Yi Wei, Shang-Hua
More informationcoverage2.ppt
Satellite Tool Kit STK/Coverage STK 82 0715 010-68745117 1 Coverage Definition Figure of Merit 2 STK Basic Grid Assets Interval Description 3 Grid Global Latitude Bounds Longitude Lines Custom Regions
More informationImproved Preimage Attacks on AES-like Hash Functions: Applications to Whirlpool and Grøstl
SKLOIS (Pseudo) Preimage Attack on Reduced-Round Grøstl Hash Function and Others Shuang Wu, Dengguo Feng, Wenling Wu, Jian Guo, Le Dong, Jian Zou March 20, 2012 Institute. of Software, Chinese Academy
More information<4D6963726F736F667420576F7264202D20B3C2B9FAD5D7C2DBCEC4C5C5B0E62E646F63>
七 年 制 专 业 学 位 论 文 论 文 题 目 正 常 人 群 上 肢 骨 与 关 节 周 围 解 剖 数 据 库 的 建 立 及 意 义 研 究 生 姓 名 指 导 教 师 姓 名 陈 国 兆 张 长 青 专 业 名 称 临 床 医 学 ( 本 硕 连 读 ) 研 究 方 向 外 科 学 ( 骨 外 科 学 ) 论 文 提 交 日 期 2013 年 6 月 正 常 人 群 上 肢 骨 与 关
More informationMicrosoft Word - p11.doc
() 11-1 ()Classification Analysis( ) m() p.d.f prior (decision) (loss function) Bayes Risk for any decision d( ) posterior risk posterior risk Posterior prob. j (uniform prior) where Mahalanobis Distance(M-distance)
More information热设计网
例 例 Agenda Popular Simulation software in PC industry * CFD software -- Flotherm * Advantage of Flotherm Flotherm apply to Cooler design * How to build up the model * Optimal parameter in cooler design
More information标题
第 46 卷 第 11 期 2 1 4 年 11 月 哈 尔 滨 工 业 大 学 学 报 JOURNAL OF HARBIN INSTITUTE OF TECHNOLOGY Vol 46 No 11 Nov 214 基 于 摄 像 测 量 法 的 在 轨 柔 性 结 构 模 态 参 数 辨 识 许 畅 王 聪 高 晶 波 张 春 芳 ( 哈 尔 滨 工 业 大 学 航 天 学 院 151 哈 尔 滨
More information****************************************************** Fundamentals of TV Tracking ****************************************************** ( ),,, :,,,,,, 1998 9 ( ISBN 7-118-01911-9),,, 1999 5 20 Email:
More information2/80 2
2/80 2 3/80 3 DSP2400 is a high performance Digital Signal Processor (DSP) designed and developed by author s laboratory. It is designed for multimedia and wireless application. To develop application
More information甄試報告1125.PDF
LabVIEW LabVIEW Laboratory Virtual Instrument Engineering Workbench G LabVIEW DAQ LabVIEW LabVIEW LabVIEW LabVIEW ph LabVIEW DAQ LabVIEW PZT LabVIEW / =2 10-8 1 LabVIEW DAQ LabVIEW DAQ DAQ LabVIEW DAQ
More informationSDS 1.3
Applied Biosystems 7300 Real-Time PCR System (With RQ Study) SDS 1.3 I. ~ I. 1. : Dell GX280 2.8GHz with Dell 17 Flat monitor 256 MB RAM 40 GB hard drive DVD-RW drive Microsoft Windows XP Operating System
More informationMicrosoft Word - 物理專文_fengli_revise_2
全 像 原 理 : 場 論 與 重 力 的 世 紀 婚 禮 文 / 林 豐 利 量 子 場 論 與 廣 義 相 對 論 是 兩 個 在 本 質 上 非 常 不 同 的 理 論 可 是 透 過 黑 洞 物 理 與 弦 論 所 啟 發 的 全 像 原 理, 卻 能 使 兩 者 互 為 一 體 的 兩 面, 增 進 了 我 們 對 量 子 重 力 的 理 解 除 此, 我 們 也 可 以 利 用 全 像
More informationuntitled
ICC 流 藍 eastwood@deepblue.com.tw 2007/01/25 數 CCD (60%) 理 更 (40%) 不見 理不, 流. 數 srgb DSLR Adobe RGB, 數 Adobe RGB 理 狀 Eye-One Match Eye-One Share Eye-One Pro 度 流 量,,, 數 數 量, 量 不 列都 精 量 io 易 Eye-One Match
More informationMicrosoft Word - A200810-897.doc
基 于 胜 任 特 征 模 型 的 结 构 化 面 试 信 度 和 效 度 验 证 张 玮 北 京 邮 电 大 学 经 济 管 理 学 院, 北 京 (100876) E-mail: weeo1984@sina.com 摘 要 : 提 高 结 构 化 面 试 信 度 和 效 度 是 面 试 技 术 研 究 的 核 心 内 容 近 年 来 国 内 有 少 数 学 者 探 讨 过 基 于 胜 任 特 征
More information<4D6963726F736F667420576F7264202D20C1B9CAB3D2A9BCE0A1B232303133A1B33536BAC520D3A1B7A232303133C4EACFC2B0EBC4EAD2A9C6B7B3E9D1E9BFECBCECB9A4D7F7CAB5CAA9B7BDB0B8B5C4CDA8D6AA2E646F63>
凉 食 药 监 2013 56 号 2013 各 县 市 食 品 药 品 监 督 管 理 局 州 食 品 药 品 检 验 所 : 现 将 凉 山 州 食 品 药 品 监 督 管 理 局 2013 年 下 半 年 药 品 抽 验 快 检 工 作 实 施 方 案 印 发 给 你 们, 请 认 真 遵 照 执 行 凉 山 州 食 品 药 品 监 督 管 理 局 2013 年 7 月 4 日 为 切 实 配
More information國立中山大學學位論文典藏.PDF
Octopus Cistopus Geomorphometric study of Octopus and Cistopus (Cephalopoda: Octopodidae) based on landmarks of beaks : : i Octopus ii iii iv v 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
More informationuntitled
識 領! 林 識 念 狀 識 力 練 力 料 年 4-6 識 行! 識 行 怒 樂!! - 1 - 怒 樂 練 狀 念 - 2 - 呂 1. 識 2. 了 3. 說 4. 5. 了 1. 2. 1. 2. 料 兩. 老 2. 3. 了 什 4. 5. 蘿 年 4~6-3 - 說 樂 & 1. 識 2. 練 3. 料 8 (... 藍 ) 1. 說 識 樂 2. 行 3. 說 說 樂 4. 不 理 敎
More informationIEEE/EIA 12207 0910023766 ??? The Frameworks Quagmire (http://www.software.org/quagmire/) () ISO 9001 ISO/IEC 12207 ISO/IEC 15504 CMM Quality System Requirements Life Cycle Processes Process Assessment
More informationVol. 22 No. 2 JOURNAL OF HARBIN UNIVERSITY OF SCIENCE AND TECHNOLOGY Apr ,,,,, Apriori,,,,,,,,
22 2 2017 4 Vol. 22 No. 2 JOURNAL OF HARBIN UNIVERSITY OF SCIENCE AND TECHNOLOGY Apr. 2017 1 2 3 2 1. 150080 2. 150080 3. 150080,,,,, Apriori,,,,,,,, ; ; ; ; DOI: 10. 15938 /j. jhust. 2017. 02. 023 TN911.
More informationMicrosoft Word - 00-巻頭言.doc
DR DR DR (6-1) DR GPSGlobal Positioning System CANCar Area Network (6-2) DR (6-3) DR - 89 - 6-1-1 GPSGlobal Positioning System 1024 768[pixel] 30[fps] GPS PC Bluetooth SSDSolid State Drive RAID 1[TB] Baumer
More informationUDC Empirical Researches on Pricing of Corporate Bonds with Macro Factors 厦门大学博硕士论文摘要库
10384 15620071151397 UDC Empirical Researches on Pricing of Corporate Bonds with Macro Factors 2010 4 Duffee 1999 AAA Vasicek RMSE RMSE Abstract In order to investigate whether adding macro factors
More informationChapter 24 DC Battery Sizing
26 (Battery Sizing & Discharge Analysis) - 1. 2. 3. ETAP PowerStation IEEE 485 26-1 ETAP PowerStation 4.7 IEEE 485 ETAP PowerStation 26-2 ETAP PowerStation 4.7 26.1 (Study Toolbar) / (Run Battery Sizing
More informationMicrosoft Word - Probability.doc
十 一 機 率 (Probability).... 分 立 變 值 (discrete variate) 及 連 續 變 值 (continuous variate)..... 連 續 變 質 (Continuous variate)/ 連 續 變 數 (Continuous variable)..... 分 立 變 值 (Discrete variate)/ 間 斷 變 數 (Discrete variable)....
More informationGH1220 Hall Switch
Unipolar Hall Switch - Medium Sensitivity Product Description The DH220 is a unipolar h all switch designed in CMOS technology. The IC internally includes a voltage regulator, Hall sensor with dynamic
More informationVol. 22 No. 4 JOURNAL OF HARBIN UNIVERSITY OF SCIENCE AND TECHNOLOGY Aug GPS,,, : km, 2. 51, , ; ; ; ; DOI: 10.
22 4 2017 8 Vol. 22 No. 4 JOURNAL OF HARBIN UNIVERSITY OF SCIENCE AND TECHNOLOGY Aug. 2017 150080 GPS,,, : 27. 36 km, 2. 51, 110. 43, ; ; ; ; DOI: 10. 15938 /j. jhust. 2017. 04. 015 U469. 13 A 1007-2683
More information精要
國 軍 主 計 財 務 軍 官 離 職 率 模 式 構 建 之 研 究 A turnover rate model for the comptroller and financial officer in the military. 張 漢 中 國 防 管 理 學 院 資 源 管 理 研 究 所 楊 承 亮 國 防 管 理 學 院 資 源 管 理 研 究 所 摘 要 一 般 探 討 離 職 的 研
More informationMicrosoft Word - 3D手册2.doc
第 一 章 BLOCK 前 处 理 本 章 纲 要 : 1. BLOCK 前 处 理 1.1. 创 建 新 作 业 1.2. 设 定 模 拟 控 制 参 数 1.3. 输 入 对 象 数 据 1.4. 视 图 操 作 1.5. 选 择 点 1.6. 其 他 显 示 窗 口 图 标 钮 1.7. 保 存 作 业 1.8. 退 出 DEFORMTM3D 1 1. BLOCK 前 处 理 1.1. 创 建
More informationMicrosoft PowerPoint - hybrid_video_coding_01.ppt
Hbrid Video Coding Principle of Hbrid Video Coding Motion-Compensated Prediction Bit Allocation Motion Models Motion Estimation Efficienc of Hbrid Video Coding Thomas Wiegand: Digital Image Communication
More informationNon-intrusive velocimetry techniques using image processing techniques
2018.10.17 ifujita@kobe-u.ac.jp 12 0.54.0m (m)
More informationPowerPoint 簡報
數 Study on the Three PCA Methods for Feature Extraction and Classification of Digital Image 羅 祿 立 94 年 9 (Principal Component Analysis, PCA) 理 料 度 更 利 PCA 益. 行 MS. 不 PCA 略理論 行 (Fe ature) 不 3. 利 類 (Bayes
More information資料HDR作1-03 HDR技術動向
作 1-3 TV Platform Div. TV Platform Div. FHD () more UHD UHD (BT.2020) [cd/m 2 ] 10 9 10 8 10 6 10 4 10 2 10 0 10-2 10-4 10-6 () ~10 5 [cd/m 2 ] 10 9 10 8 10 6 10 4 10 2 10 0 10-2 10-4 10-6 2014 XDR/XDR-Pro
More information第一章
課 程 名 稱 : 光 纖 傳 輸 實 務 與 實 習 1. 課 程 概 述 : 光 纖 傳 輸 實 務 與 實 習 為 隔 年 開 授 之 課 程, 此 高 等 課 程 實 習 項 目 之 內 容 較 具 彈 性, 以 教 導 學 生 如 何 使 用 設 計 工 具 與 發 揮 設 計 能 力 為 目 標 新 編 了 光 纖 光 放 大 器 模 擬 設 計 實 習 教 材, 包 含 摻 鉺 光 纖
More information衛星影像分類
年 理 理 立 立 列 SPOT 立 年 理 2 , 量 不 料 -Raster 量,, -Vector? 立 年 理 3 料 (Binary Data) 省 理 率 料來 CCD (Charge Couple Device) (Scanner) 數, 數 錄? 立 年 理 4 :Picture Element or Pixel : 不 不 狀 X,Y Column,Row Sample,Line
More information摘 要 本 校 多 媒 體 設 計 系 與 上 海 戲 劇 學 院 創 意 學 院 在 多 次 聯 繫 交 流 之 下, 已 簽 署 合 作 備 忘 錄, 積 極 尋 求 兩 校 合 作 教 學 與 共 同 創 作 之 機 會 藉 由 本 系 學 生 作 品 腦 殘 公 寓 入 圍 第 五 屆 中
出 國 報 告 ( 出 國 類 別 : 其 他 - 國 際 競 賽 ) 中 國 獨 立 遊 戲 節 國 際 學 生 競 賽 及 上 海 戲 劇 學 院 合 作 洽 談 服 務 機 關 : 國 立 臺 中 科 技 大 學 多 媒 體 設 計 系 姓 名 職 稱 : 金 啟 平 專 案 講 師 派 赴 國 家 : 中 國 出 國 期 間 :102.9.14 102.9.17 報 告 日 期 :102.11.20
More information108 臺 中 教 育 大 學 學 報 : 人 文 藝 術 類 Abstract To satisfy the audiences' pursuing of image realism and content imagination, the combination of live footage
107 臺中教育大學學報 人文藝術類 00 年 ( ) 0 - 使用攝影機追蹤的實景合成電腦動畫創作研究 Applying Camera Tracking as a Tool to Composite Computer Animation onto Video Editing 呂洽毅 鐘世凱 Cha-Yi Lu Shih-Kai Chung 收件日期 年 月 日 接受日期 年 月 日 摘 要 為了滿足閱聽者對影像真實感與想像力的需求
More informationLK110_ck
Ck 电子琴 LK110CK1A Ck-1 1. 2. 1. 2. 3. (+) ( ) Ck-2 1. 2. 3. * 1. 2. 3. Ck-3 Ck-4 LCD LCD LCD LCD LCD LCD 15 * * / MIDI Ck-5 100 50 100 100 100 1 2 MIDI MIDI Ck-6 ... Ck-1... Ck-6... Ck-8... Ck-9... Ck-10...
More informationCOCO18-DensePose-BUPT-PRIV
* Beijing University of Posts and Telecommunications (BUPT) COCO 2018 DensePose Test AP AP50 AP75 APm APl BUPT-PRIV (Ours) 64 92 75 57 67 PlumSix 58 89 66 50 61 ML_Lab 57 89 64 51 59 Sound of silent 57
More informationMicrosoft PowerPoint - ATF2015.ppt [相容模式]
Improving the Video Totalized Method of Stopwatch Calibration Samuel C.K. Ko, Aaron Y.K. Yan and Henry C.K. Ma The Government of Hong Kong Special Administrative Region (SCL) 31 Oct 2015 1 Contents Introduction
More informationMicrosoft PowerPoint - Aqua-Sim.pptx
Peng Xie, Zhong Zhou, Zheng Peng, Hai Yan, Tiansi Hu, Jun-Hong Cui, Zhijie Shi, Yunsi Fei, Shengli Zhou Underwater Sensor Network Lab 1 Outline Motivations System Overview Aqua-Sim Components Experimental
More informationSep (SCI) 10. Jiann-Ming Wu, Annealing by two sets of interactive dynamics, IEEE Trans. on Systems Man and Cybernetics Part B-Cybernetics 34 (3)
03-863-3531 E-mail jmwu@mail.ndhu.edu.tw /(1990/06-1994/06) (1988/06-1990/06) (1984/09-1988/06) / (1997/8-) (1996/08-1997/07) () 1996 8 Wu, Jiann-Ming ( SCI EI SSCI TSSCI EconLit A&HCI ) 1. Pei-Hsun Hsu
More information[9] R Ã : (1) x 0 R A(x 0 ) = 1; (2) α [0 1] Ã α = {x A(x) α} = [A α A α ]. A(x) Ã. R R. Ã 1 m x m α x m α > 0; α A(x) = 1 x m m x m +
2012 12 Chinese Journal of Applied Probability and Statistics Vol.28 No.6 Dec. 2012 ( 224002) Euclidean Lebesgue... :. : O212.2 O159. 1.. Zadeh [1 2]. Tanaa (1982) ; Diamond (1988) (FLS) FLS LS ; Savic
More informationVX2753mh-LED-1_UG_KRN.book
ViewSonic VX2753mh-LED LCD Display Model No. : VS13918 ...1...2 RoHS...3...4...4...5...5...6 LCD...7 OSD...7...8...10...15...16...17 LCD...18...19 ViewSonic VX2753mh-LED FCC FCC 15.. (1), (2). FCC 15 B..,,..
More information% % 34
* 2000 2005 1% 1% 1% 1% * VZDA2010-15 33 2011. 3 2009 2009 2004 2008 1982 1990 2000 2005 1% 1 1 2005 1% 34 2000 2005 1% 35 2011. 3 2000 0. 95 20-30 209592 70982 33. 9% 2005 1% 258 20-30 372301 115483 31.
More informationuntitled
1 1. 1 1. 1. 1 image I(x,y,z,t) object to image 1. 1. 2 1 X Rays, Gamma-Rays, X Rays CT 1. 1. 3 INPUT Processing OUTPUT 1. 1. 4 Anolog Image Processing 2 25 / 1. 1. 5 Digital Image Processing Digitizing,
More informationMicrosoft PowerPoint - NCBA_Cattlemens_College_Darrh_B
Introduction to Genetics Darrh Bullock University of Kentucky The Model Trait = Genetics + Environment Genetics Additive Predictable effects that get passed from generation to generation Non-Additive Primarily
More informationSVM OA 1 SVM MLP Tab 1 1 Drug feature data quantization table
38 2 2010 4 Journal of Fuzhou University Natural Science Vol 38 No 2 Apr 2010 1000-2243 2010 02-0213 - 06 MLP SVM 1 1 2 1 350108 2 350108 MIP SVM OA MLP - SVM TP391 72 A Research of dialectical classification
More information11第十一章階層線性模式.DOC
11.1 11.1.1 (student-level) (personal-level) ( ) (school-level) (organization-level) ( ) 1. (disaggregation) (estimated standard errors) (type one error). (aggregation) (within-group) (1997) (hierarchical
More information投影片 1
Coherence ( ) Temporal Coherence Michelson Interferometer Spatial Coherence Young s Interference Spatiotemporal Coherence 參 料 [1] Eugene Hecht, Optics, Addison Wesley Co., New York 2001 [2] W. Lauterborn,
More informationValue Chain ~ (E-Business RD / Pre-Sales / Consultant) APS, Advanc
Key @ Value Chain fanchihmin@yahoo.com.tw 1 Key@ValueChain 1994.6 1996.6 2000.6 2000.10 ~ 2004.10 (E- RD / Pre-Sales / Consultant) APS, Advanced Planning & Scheduling CDP, Collaborative Demand Planning
More informationuntitled
01 1-1 1-2 1-3 Chapter 1-1 (statistics) (uncertainty) 1. (1) (descriptive statistics) (2) (inferential statistics) (statistical inference) (inductive statistics) (parametric statistics) (nonparametric
More information34 www. cjig. cn wavelet transform 1 2 JPEG LIVE E s o = 1 T Σ log 2 C s o + 1 E T C s o Lu Wen contourlet C 0 7 N
17 1 2012 1 Journal of Image and Graphics Vol. 17No. 1 Jan. 2012 TP391. 4 A 1006-8961201201-0033-07. NSS J. 2012 171 33-39 NSS 214122 NSS No-reference image quality assessment based on natural scene statistics
More informationBC04 Module_antenna__ doc
http://www.infobluetooth.com TEL:+86-23-68798999 Fax: +86-23-68889515 Page 1 of 10 http://www.infobluetooth.com TEL:+86-23-68798999 Fax: +86-23-68889515 Page 2 of 10 http://www.infobluetooth.com TEL:+86-23-68798999
More informationPowerPoint Presentation
Decision analysis 量化決策分析方法專論 2011/5/26 1 Problem formulation- states of nature In the decision analysis, decision alternatives are referred to as chance events. The possible outcomes for a chance event
More information國家圖書館典藏電子全文
I Abstract II III ... I Abstract...II...III... IV... VI 1...1 2...3 2-1...3 2-2...4 2-3...6 2-4...6 3...8 3-1...8 3-2...10 4...12 5...15 5-1...15 5-2...17 IV 5-3...18 6...21 6-1...21 6-2...22 6-3...22
More informationAbstract Due to the improving of living standards, people gradually seek lighting quality from capacityto quality. And color temperature is the important subject of it. According to the research from aboard,
More informationiml88-0v C / 8W T Tube EVM - pplication Notes. IC Description The iml88 is a Three Terminal Current Controller (TTCC) for regulating the current flowi
iml88-0v C / 8W T Tube EVM - pplication Notes iml88 0V C 8W T Tube EVM pplication Notes Table of Content. IC Description.... Features.... Package and Pin Diagrams.... pplication Circuit.... PCB Layout
More information网络“超常”形式单位认知研究与规范
汉 语 复 合 词 超 常 组 合 的 认 知 研 究 尹 铂 淳 ( 湖 南 师 范 大 学 外 国 语 学 院, 湖 南 长 沙 410081) 内 容 摘 要 : 汉 语 复 合 词 超 常 组 合 是 相 对 于 常 规 组 合 而 言 的, 是 人 类 思 维 创 新 的 体 现 经 历 过 超 常 组 合 的 汉 语 复 合 词 拥 有 令 人 难 以 准 确 把 握 的 词 义, 欲
More information(156) / Spurious Regression Unit Root Test Cointergration TestVector Error Correction Model Granger / /
(155) * ** / / / / 1973 ~1974 1979 ~1980 1987 ~1989 * ** (156) 1990 2004 1997 1996 1980 / Spurious Regression Unit Root Test Cointergration TestVector Error Correction Model Granger / / (157) Hedonic Price
More information彩色地图中道路的识别和提取
9310016, i ii Abstract This thesis is on the researching of recognizing the roads in map image by computer. Based on the theory of Pattern Recognition, there is a method to be discussed, which can recognize
More information08陈会广
第 34 卷 第 10 期 2012 年 10 月 2012,34(10):1871-1880 Resources Science Vol.34,No.10 Oct.,2012 文 章 编 号 :1007-7588(2012)10-1871-10 房 地 产 市 场 及 其 细 分 的 调 控 重 点 区 域 划 分 理 论 与 实 证 以 中 国 35 个 大 中 城 市 为 例 陈 会 广 1,
More information6-7 6-8 6-9 Process Data flow Data store External entity 6-10 Context diagram Level 0 diagram Level 1 diagram Level 2 diagram 6-11 6-12
6-1 6-2 6-3 6-4 6-5 6-6 6-7 6-8 6-9 Process Data flow Data store External entity 6-10 Context diagram Level 0 diagram Level 1 diagram Level 2 diagram 6-11 6-12 6-13 6-14 6-15 6-16 6-17 6-18 6-19 6-20 6-21
More informationa b c d e f g C2 C1 2
a b c d e f g C2 C1 2 IN1 IN2 0 2 to 1 Mux 1 IN1 IN2 0 2 to 1 Mux 1 Sel= 0 M0 High C2 C1 Sel= 1 M0 Low C2 C1 1 to 2 decoder M1 Low 1 to 2 decoder M1 High 3 BCD 1Hz clk 64Hz BCD 4 4 0 1 2 to 1 Mux sel 4
More informationPowerPoint Presentation
Linear Progamming- the Simple method with greater-than-or-equal-to or equality minimization problem Quantitative deciion making technique /5/6 Tableau form- dealing with greaterthan-or-equal-to contraint
More informationMicrosoft PowerPoint - STU_EC_Ch08.ppt
樹德科技大學資訊工程系 Chapter 8: Counters Shi-Huang Chen Fall 2010 1 Outline Asynchronous Counter Operation Synchronous Counter Operation Up/Down Synchronous Counters Design of Synchronous Counters Cascaded Counters
More information2782_OME_KM_Cover.qxd
数码说明书之家 2005.09.06 www.54gg.com 2 3 4 5 6 7 9 8...14...14...17...18...19...20...20...20...21...22...23...24...24...25...26...28...28...29...29...30...32...32 EVF LCD...32...33...34...34...35...35...36...36...37...38...39...40...40...41...41...42...43...44...45...45...46...47...48...49...50...50
More information朝 陽 科 技 大 學 2015 年 工 業 設 計 系 專 題 設 計 報 告 書 麵 對 麵 - 中 西 麵 食 料 理 器 具 設 計 指 導 教 授 : 劉 哲 揚 設 計 者 : 翁 苡 恬 中 華 民 國 一 0 四 年 六 月 二 日 麵 對 麵 - 中 西 麵 食 料 理 器 具 設
朝 陽 科 技 大 學 2015 年 工 業 設 計 系 專 題 設 計 報 告 書 麵 對 麵 - 中 西 麵 食 料 理 器 具 設 計 指 導 教 授 : 劉 哲 揚 設 計 者 : 翁 苡 恬 中 華 民 國 一 0 四 年 六 月 二 日 麵 對 麵 - 中 西 麵 食 料 理 器 具 設 計 1 目 錄 壹 設 計 摘 要... 7 貳 設 計 說 明... 8 一 設 計 動 機...
More informationK-means
zwp@ustc.edu.cn Office: 1006 Phone: 63600565 http://staff.ustc.edu.cn/~zwp/ http://fisher.stat.ustc.edu.cn 1.1....................... 1 1.2............... 6 1.3.................... 11 1.3.1...............
More informationV6800/V6600 3D
V6800/V6600 3D V6600/V6800 3D R 2000 2 3 4 5 R 6 7 8 The VIP (Video Interface Port) Connector are used for third party add-on modules, such as video capture cards or television tuners. DDR: Double Data
More informationStochastic Processes (XI) Hanjun Zhang School of Mathematics and Computational Science, Xiangtan University 508 YiFu Lou talk 06/
Stochastic Processes (XI) Hanjun Zhang School of Mathematics and Computational Science, Xiangtan University hjzhang001@gmail.com 508 YiFu Lou talk 06/04/2010 - Page 1 Outline 508 YiFu Lou talk 06/04/2010
More informationPin Configurations Figure2. Pin Configuration of FS2012 (Top View) Table 1 Pin Description Pin Number Pin Name Description 1 GND 2 FB 3 SW Ground Pin.
Features Wide 3.6V to 32V Input Voltage Range Output Adjustable from 0.8V to 30V Maximum Duty Cycle 100% Minimum Drop Out 0.6V Fixed 300KHz Switching Frequency 12A Constant Output Current Capability Internal
More information一 课 程 负 责 人 情 况 姓 名 吴 翊 性 别 男 出 生 年 月 1948.08.28 1. 基 本 信 息 学 位 硕 士 职 称 教 授 职 务 所 在 院 系 理 学 院 数 学 与 系 统 科 学 系 电 话 13808485157 研 究 方 向 数 据 处 理 近 三 年 来
湖 南 省 普 通 高 等 学 校 省 级 精 品 课 程 复 核 申 报 表 学 校 名 称 : 课 程 名 称 : 课 程 负 责 人 : 立 项 时 间 : 国 防 科 技 大 学 概 率 论 与 数 理 统 计 吴 翊 2009 年 课 程 网 站 地 址 :jpkc2009.nudt.edu.cn/gllysltj jpkc.nudt.edu.cn/gltj 湖 南 省 教 育 厅 制 二
More informationMonetary Policy Regime Shifts under the Zero Lower Bound: An Application of a Stochastic Rational Expectations Equilibrium to a Markov Switching DSGE
Procedure of Calculating Policy Functions 1 Motivation Previous Works 2 Advantages and Summary 3 Model NK Model with MS Taylor Rule under ZLB Expectations Function Static One-Period Problem of a MS-DSGE
More information城 市 發 展 City 學 術 論 壇 摘 要 高 雄 市 近 年 來 關 注 城 市 意 的 議 題, 在 政 治 部 門 有 計 畫 的 推 行, 文 化 創 意 產 業 的 推 波 助 瀾, 及 西 方 城 市 文 化 的 影 響 市 民 意 識 的 興 起 與 廣 泛 的 迴 響 下, 企 圖 營 造 出 高 雄 市 自 己 的 獨 特 風 格, 設 計 和 規 劃 城 市 的 整 體 意,
More informationThesis for the Master degree in Engineering Research on Negative Pressure Wave Simulation and Signal Processing of Fluid-Conveying Pipeline Leak Candi
U17 10220 UDC624 Thesis for the Master degree in Engineering Research on Negative Pressure Wave Simulation and Signal Processing of Fluid-Conveying Pipeline Leak Candidate:Chen Hao Tutor: Xue Jinghong
More information( ) t ( ) ( ) ( ) ( ) ( ) t-
(Statistics). (Descriptive Statistics). (Inferential Statistics) (Inductive Statistics) ( ) t ( ) ( ) ( ) ( ) ( ) t- ( ) ( ) ( )? ( ) ( )? ( ) )?( t ) ( )? ( ) ( ) ( ) ( ) ( ) ( )? ( ) ( ) ( )? ( )?( t
More informationPowerPoint Presentation
ITM omputer and ommunication Technologies Lecture #4 Part I: Introduction to omputer Technologies Logic ircuit Design & Simplification ITM 計算機與通訊技術 2 23 香港中文大學電子工程學系 Logic function implementation Logic
More information)
.. 1. 2. ) () () Pilot test () 1. 2. 3. 4. Scale (1). (nominal scale) 1. 2. 3. (1,2,3) (scale value) (arithmetic mean) (median) (mode) (chi-square test) (2). (ordinal scale) 5 1 A>B>C>D>E A B C D (non-parametric
More information前言
球 形 屋 頂 建 築 簇 群 所 受 表 面 風 壓 特 性 研 究 陳 若 華 建 國 科 技 大 學 空 間 設 計 系 副 教 授 摘 要 本 研 究 利 用 球 形 屋 頂 建 築 物 氣 動 力 模 型, 於 風 洞 中 量 測 其 所 受 表 面 風 壓, 風 壓 量 測 利 用 電 子 式 壓 力 掃 瞄 器 量 測 建 築 物 表 面 同 步 的 風 壓 力 資 料, 量 測 的
More informationText 文字输入功能 , 使用者可自行定义文字 高度, 旋转角度 , 行距 , 字间距离 和 倾斜角度。
GerbTool Wise Software Solution, Inc. File New OPEN CLOSE Merge SAVE SAVE AS Page Setup Print Print PreView Print setup (,, IMPORT Gerber Wizard Gerber,Aperture Gerber Gerber, RS-274-D, RS-274-X, Fire9000
More informationuntitled
6 20 90 BellCore Ethernet variable bit rate VBR fractal self-similarity 994 IEEE/ACM Transactions on Networking On the self-similarity nature of Ethernet traffic extended version LAN WAN CCSN/SS7 ISDN
More informationConcept of Hyper Kamiokande (20 times Super K) 48m x 50m x 250m x 2 Total mass ~ 1 Mton Photocathode coverage ~40% of surface ~200,000 PMTs => prohibi
R&D of a Large Format Hybrid Photo-Detector (HPD) for a Next Generation Water Cherenkov Detector Tokyo - HPK joint R&D program H.Aihara University of Tokyo HPK =Hamamatsu Photonics 1 presented at Next
More information國立中山大學學位論文典藏.PDF
Air Visibility Forecasting via Artificial Neural Networks and Feature Selection Techniques 2003 7 Air Visibility Forecasting via Artificial Neural Networks and Feature Selection Techniques Sequential Floating
More informationMicrosoft Word - LD5515_5V1.5A-DB-01 Demo Board Manual
Subject LD5515 Demo Board Model Name (5V/1.5A) Key Features Built-In Pump Express TM Operation Flyback topology with PSR Control Constant Voltage Constant Current High Efficiency with QR Operation (Meet
More informationiml v C / 0W EVM - pplication Notes. IC Description The iml8683 is a Three Terminal Current Controller (TTCC) for regulating the current flowin
iml8683-220v C / 0W EVM - pplication Notes iml8683 220V C 0W EVM pplication Notes Table of Content. IC Description... 2 2. Features... 2 3. Package and Pin Diagrams... 2 4. pplication Circuit... 3 5. PCB
More information, : III
河北师范大学硕士学位论文石家庄公共卫生间设计研究姓名 : 武秀娥申请学位级别 : 硕士专业 : 美术学指导教师 : 张爱民 20100406 , : III Abstract Rapid economic development for the city's construction and development, cities around the country to the city in
More informationiml v C / 4W Down-Light EVM - pplication Notes. IC Description The iml8683 is a Three Terminal Current Controller (TTCC) for regulating the cur
iml8683-220v C / 4W Down-Light EVM - pplication Notes iml8683 220V C 4W Down Light EVM pplication Notes Table of Content. IC Description... 2 2. Features... 2 3. Package and Pin Diagrams... 2 4. pplication
More informationMicrosoft PowerPoint - CH 04 Techniques of Circuit Analysis
Chap. 4 Techniques of Circuit Analysis Contents 4.1 Terminology 4.2 Introduction to the Node-Voltage Method 4.3 The Node-Voltage Method and Dependent Sources 4.4 The Node-Voltage Method: Some Special Cases
More information